Experiences with Sweep3D implementations in Co-array Fortran

Abstract

As part of the recent focus on increasing the productivity of parallel application developers, Co-array Fortran (CAF) has emerged as an appealing alternative to the Message Passing Interface (MPI). CAF belongs to the family of global address space parallel programming languages; such languages provide the abstraction of globally addressable memory accessed using one-sided communication. At Rice University we are developing cafc, an open source, multiplatform CAF compiler. Our earlier studies show that cafc-compiled CAF programs achieve similar performance to that of corresponding MPI codes for the NAS Parallel Benchmarks. In this paper, we present a study of several CAF implementations of Sweep3D on four modern architectures. We analyze the impact of using one-sided communication in Sweep3D, identify potential sources of inefficiencies and suggest ways to address them. Our results show that we achieve comparable performance to that of the MPI version on three cluster-based architectures and outperform it by up to 10% on the SGI Altix 3000.

Multithreaded global address space communication techniques for gyrokinetic fusion applications on ultra-scale platforms

“…GTS production runs will be conducted in a hybrid distributed / shared memory mode with a maximum number of OpenMP threads per NUMA node because of optimal memory utilization and performance. Hence, scaling studies of our CAF algorithm and the MPI shifter implementations run with a constant setup of 6 OpenMP threads per instance 7 . In addition, we are interested in the performance characteristics in regard to varying number of OpenMP threads (i.e., keeping the number of instances constant per experiment) to analyze the quality of the OpenMP support and to gain insights for further optimization potential concerning software and hardware.…”

“…The CAF 2.0 runtime uses the GASNet library [4] for one-sided communication. Applications of Coarray Fortran to the NAS parallel benchmarks (Coarfa et al [6]) and to the Sweep3D neutron transport benchmark (Coarfa et al [7]) show nearly equal or slightly better performance than their MPI counterparts.…”

Aerothermal Load and Drag Force Analysis of the Electromagnetically Launched Projectiles Under Rarefied Gas Conditions

“…Developing parallel programs with CAF can be achieved with less lines of code without sacrificing performance. The parallel performance of the CAF is comparable with the MPIs performance [14]. CAF is a parallel programming language, which is easy to read, understand, and maintain compared with other parallelization tools.…”

Aerothermodynamics analysis of the electromagnetically launched projectiles under rarefied gas conditions

“…Developing parallel programs with CAF can be achieved with less lines of code without sacrificing performance. The parallel performance of the CAF is comparable to MPI's performance [13]. Shortly, it can be said that CAF is a parallel programming language which is easy to read, understand, and maintain comparing to other parallelization tools.…”